CN106248368B - Combustion engine turbine blade fault detection method based on deep learning - Google Patents
Combustion engine turbine blade fault detection method based on deep learning Download PDFInfo
- Publication number
- CN106248368B CN106248368B CN201610839954.0A CN201610839954A CN106248368B CN 106248368 B CN106248368 B CN 106248368B CN 201610839954 A CN201610839954 A CN 201610839954A CN 106248368 B CN106248368 B CN 106248368B
- Authority
- CN
- China
- Prior art keywords
- points
- signal
- turbine blade
- maximum value
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N25/00—Investigating or analyzing materials by the use of thermal means
- G01N25/72—Investigating presence of flaws
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Theoretical Computer Science (AREA)
- Biomedical Technology (AREA)
- Evolutionary Computation (AREA)
- Pathology (AREA)
- Biochemistry (AREA)
- Artificial Intelligence (AREA)
- Analytical Chemistry (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Immunology (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Chemical & Material Sciences (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
- Control Of Turbines (AREA)
Abstract
The invention belongs to the field of fault detection of gas turbines, and particularly relates to a deep learning-based fault detection method for turbine blades of a gas turbine. The invention comprises the following steps: (1) preprocessing temperature data of the turbine blade; (2) extracting a turbine blade temperature characteristic vector; (3) fault diagnosis based on a deep learning network; (4) turbine blade fault detection for a combustion engine. Aiming at the problem that the temperature acquisition data sample of the turbine blade is large, the deep learning method is introduced into the fault diagnosis of the turbine blade of the gas turbine for the first time, so that the diversity development of the fault diagnosis of the turbine blade of the gas turbine is promoted, and the accuracy of fault detection is improved.
Description
Technical Field
The invention belongs to the field of fault detection of gas turbines, and particularly relates to a deep learning-based fault detection method for turbine blades of a gas turbine.
Background
Gas turbine engines, also known as gas turbines, which are a form of heat engine, have been used in the last 30 centuries for large military and civilian applications such as aviation and ships, and have been regarded as an important manifestation of the national comprehensive strength since the invention. Since 1960, the gas turbine comprehensively replaces a piston engine, the gas turbine is applied suddenly and violently, and the gas turbine also becomes irreplaceable in a generator and a propulsion system of a tank in each country. Because the device has the excellent characteristics of small volume, light weight, less pollution, high reliability, outstanding cost performance, extremely high efficiency and the like, the device reflects the utilization value of the device again under the large-premise environment of improving the environment and adjusting the energy industry mechanism since the 21 st century, so far, the industrial scale of the device becomes larger and larger, and the product application becomes wider and wider.
The main structure of the gas turbine mainly comprises a combustion chamber, a gas compressor, a turbine, a tail nozzle and other components, and can be divided into the following parts according to classification: the turbo jet engine, the turbo fan engine, the turboprop engine, the turbine bearing engine and the like, wherein the turbo fan engine is most widely applied, and the aviation field, the ship field and the generator system mostly adopt the turbo fan engine as a preferred scheme for providing power. The gas turbine industry in China starts late, but develops rapidly, and through years of scientific research and development, China makes great progress in the aspects of independent innovation of the gas turbine, product development, digestion and absorption and foreign advanced technology. Meanwhile, due to the adjustment of the structure of the energy industry, the yield of the natural gas is continuously increased, and good conditions and environment are provided for the development of the gas turbine taking the natural gas as the fuel. According to the plan of national development and improvement committee, the total installed quantity of the gas turbine in China reaches 5500 ten thousand kilowatts as far as 2020, and meanwhile, the gas turbine becomes the most huge gas turbine development market in the world and has infinite prospect.
Deep learning is an artificial intelligence mode proposed by Hinton et al in 2006, which can abstract low-level features into high-level features closer to the nature of the features, has been successfully implemented in handwriting recognition, speech recognition, face recognition and the like, and is a hot point of current research. The principle of the traditional machine learning method such as hidden Markov model, support vector machine, maximum entropy model and the like in processing feature vectors is to map the original input features into feature intervals, so that the feature structure becomes simpler and has certain limitation. However, these methods tend to be frustrating when faced with the classification of complex problems when sample capacity and computational power are limited. In contrast, the deep learning mode can deal with the problem of extracting essential features of a data set from a large-capacity sample, effectively approximates a complex function, and classifies input data after being processed.
Disclosure of Invention
The invention aims to provide a turbine blade fault detection method of a combustion engine based on deep learning, which can improve the fault detection fault rate.
The purpose of the invention is realized as follows:
(1) the temperature data preprocessing of the turbine blade comprises the following steps:
(1.1) periodic Signal extraction
The turbine engine analyzed consisted of 86 blades, the data for radiation thermometry was obtained after a number of cycles of rotation of all the blades,
according to the discriminant formula of the periodic property: f (x) ═ f (x + T)
Wherein T is the period of the function, and whether the temperature data is period data is judged by the above formula; the minimum value and the maximum value are taken as the discrimination standard to carry out the period discrimination,
taking a certain maximum value point as a starting point, counting all the maximum value points in a certain section of interval, calling the points as test points, obtaining two adjacent test points with the same temperature, then respectively sampling backwards and comparing the temperatures of the sampling points, and if the two adjacent test points with the same temperature are the same, determining that the period is one;
(1.2) variable Condition temperature data distribution
The working conditions are divided into the following three working conditions: 0.6 operating mode, 0.8 operating mode and 1.0 operating mode, the turbine blade rotational speed under three kinds of operating modes respectively is: 8635/rpm, 8920/rpm, 9138/rpm;
(1.3) Single blade temperature data distribution
Establishing a segmentation algorithm for single blade temperature data on extreme value searching, counting maximum value points in the data, comparing the temperature data of each extreme value point, and selecting the point with the maximum temperature data, namely the maximum value point, and marking as a0With a0Searching a period backwards as a starting point, namely 40 sampling points, and recording a maximum value point in the period; making a judgment on the maximum value points, when the maximum value points cannot be larger than 10% of the adjacent maximum value points, if the maximum value points are larger than 10% of the adjacent maximum value points, changing the maximum value points into the starting points a0The process is circulated for 86 times to obtain the period of 86 blades, the sequence numbers from 01 to 86 are carried out on the blades, and the extracted features are used for state classification detection;
(2) extracting turbine blade temperature feature vectors
Extracting the temperature characteristic vector of the blade by adopting a method based on ensemble empirical mode decomposition; changing the number of extreme points and the distribution interval of the extreme points of the signal or data by adding different Gaussian noises with given amplitudes into the signal every time, and then carrying out overall average cancellation on IMF components obtained by multiple decomposition to cancel the noises added into the signal;
(2.1) initializing a noise amplitude value added to the signal;
(2.2) decomposing the ith added noise signal;
(2.2.1) adding a Gaussian signal n with a certain amplitude to the signal x (t)i(t),xi(t)=ni(t) + x (t), where the left side of the equation represents the signal formed after the i-th addition of the noise signal; the first term on the right of the equation represents the gaussian noise signal; x (t) represents the original signal;
(2.2.2) for xi(t) EMD decomposition to obtain a set of IMF components ci(I ═ 1,2, 3.., I), where c isiRepresenting IMF components obtained by the ith decomposition; i represents the number of decompositions;
(2.2.3) if I < I, returning to step (2.2.1) and so that I ═ I +1, repeating step (2.2.1) and step (2.2.2) until I ═ I;
(2.3) calculating the overall average value by the I decomposition, denoted by the symbol y;
(2.4) saving the total average value of all IMF components as a final IMF component;
(3) fault diagnosis based on deep learning network
Selecting a restricted Boltzmann machine model as a depth network design model, adjusting the restricted Boltzmann machine model through the progressive between layers, forming a transmission layer of the whole depth model by using 1 hidden layer and 1 visual output layer, and initializing the weight of the whole network;
the restricted Boltzmann model is expressed as RBM (W, b, c, v)0) Where W is the weight connection matrix between the first hidden layer and the second hidden layer, b is the bias of the hidden layer, c is the bias of the input layer, v0Represents a sample set used in training;
(3.1) for all nodes i of the hidden layer, calculating mapping operation between the hidden layersI.e. P (h)0i=1|v0) According to P (h)0i=1|v0) Sampling to obtain h0iSigm () is a mapping function;
(3.2) for all nodes j of the visual layers, calculating the mapping operation between the visual layersI.e. P (v)1j=1|h0) According to P (v)1j=1|h0) Sampling to obtain v1j;
(3.3) for the used hidden layer node i, calculating the mapping operation between the hidden layer and the hidden layerI.e. P (h)1i=1|v1) According to P (h)1i=1|v1) Updating the bias parameters of the connection weight; the update formula is as follows:
W=W-ε(h0v0'-Q(h1=1|v1)v1')
b=b-ε(h0-Q(h1=1|v1))
c=c-ε(v0-v1)
(4) detecting faults of turbine blades of a combustion engine: constructing a gas turbine blade fault detection system for an algorithm experiment, simulating 6 types of fault data, 10 groups of samples of each type, and 60 groups of data, carrying out EEMD decomposition on a simulated turbine blade fault signal, then solving the energy of IMF of the simulated turbine blade fault signal, selecting the first 8 components, enabling the capacity of a residual item to be zero, and then solving a feature vector; dividing the data into two parts, randomly taking 48 groups as training samples, and taking the other 12 groups as test samples; training the deep learning network by using the training samples; and classifying the samples to be tested by using the trained deep learning network, and outputting a classification result.
The invention has the beneficial effects that: aiming at the problem that the temperature acquisition data sample of the turbine blade is large, the deep learning method is introduced into the fault diagnosis of the turbine blade of the gas turbine for the first time, so that the diversity development of the fault diagnosis of the turbine blade of the gas turbine is promoted, and the accuracy of fault detection is improved.
Drawings
FIG. 1 is a block diagram of turbine blade fault detection based on deep learning;
FIG. 2 turbine blade segmentation flow chart.
Detailed Description
The invention is further described below with reference to the accompanying drawings.
The invention provides a turbine blade fault detection method of a combustion engine based on deep learning, which comprises the steps of preprocessing turbine blade temperature data, including periodic signal extraction and segmentation, different working condition temperature data analysis, single blade temperature data distribution research, then extracting the characteristics of a turbine blade, and extracting the characteristic vector of the turbine blade by adopting a method based on Ensemble Empirical Mode Decomposition (EEMD). And finally, carrying out turbine blade fault diagnosis based on the deep learning network. Compared with a BP neural network and a support vector machine, the fault diagnosis accuracy is higher.
The invention relates to a method for detecting faults of turbine blades of a gas turbine, which comprises the steps of analyzing and preprocessing temperature data of the turbine blades, extracting periodic signals and analyzing data under different working conditions, segmenting the temperature data of single turbine blades, extracting temperature characteristic vectors of the turbine blades, and then detecting the faults of the turbine blades of the gas turbine based on a deep learning theory.
The invention relates to a method for detecting faults of turbine blades of a gas turbine, which comprises the steps of preprocessing temperature data of the turbine blades to obtain data of a single turbine blade in one period under different working conditions, segmenting the temperature data of the single turbine blade, extracting a temperature characteristic vector of the turbine blade, and then detecting the faults of the turbine blades of the gas turbine based on a BP artificial network adopting a depth model.
The method comprises the following concrete implementation steps:
(1) the temperature data preprocessing of the turbine blade comprises the following steps:
extraction of periodic signal
The turbine engine analyzed consisted of 86 blades and the data for radiation thermometry was obtained after a number of cycles of rotation of all the blades.
According to the discriminant formula of the periodic property: f (x) ═ f (x + T)
Where T is the period of the function, whether these temperature data are period data is determined by the above equation. And (3) taking the minimum value and the maximum value as discrimination standards to carry out period judgment, and finding out that almost the same waveform appears at intervals of a sampling interval, so that the acquired temperature data is judged to be period data.
Taking a certain maximum value point as a starting point, counting all the maximum value points in a certain section of interval, calling the points as 'test points', obtaining two adjacent test points with the same temperature, then respectively sampling backwards and comparing the temperatures of the sampling points, and if the two adjacent test points with the same temperature are the same, determining that the period is the period.
Operating mode-changing temperature data distribution
The rotating speed of the turbine is fast when the turbine runs, so that the specific moment when the fault occurs cannot be determined, and the rotating speed condition of the turbine engine at a certain temperature in the working process within a certain period of time when the fault occurs can only be determined. The working conditions are divided into the following three working conditions: 0.6 operating mode, 0.8 operating mode and 1.0 operating mode, the turbine blade rotational speed under three kinds of operating modes respectively is: 8635/rpm, 8920/rpm, 9138/rpm.
Distribution of single blade temperature data
In turbine engines, there is an overlap between blades, and therefore, the temperature data needs to be further analyzed to identify specific blade positions. Establishing a segmentation algorithm for single blade temperature data on extreme value searching, counting extreme values in the data by taking a maximum value as an example, comparing the temperature data of all the extreme values, and selecting a point with the maximum temperature data, namely a maximum value point, recorded as a0With a0For the starting point, a cycle, i.e. 40 sample points, is searched backwards, and the maximum point within the cycle is recorded. Making a decision on the maxima points when the maxima points cannot be greater than 10% of their neighbors, e.g.If it is more than 10% of the neighboring points, the maximum point is changed to the starting point a0The method is circulated for 86 times, the period of 86 blades is obtained, the 01 to 86 sequence numbers are carried out on the blades, and the characteristics of the blades can be extracted for state classification detection.
(2) Turbine blade temperature feature vector extraction
The extraction of the blade temperature feature vector is performed by a method based on Ensemble Empirical Mode Decomposition (EEMD). EEMD is a signal analysis method which assists analysis through noise, the method utilizes the characteristic that white Gaussian noise has uniform frequency distribution to solve the modal aliasing phenomenon caused by Decomposition of an Empirical Mode Decomposition (EMD) method, changes the number of extreme points and the distribution interval of the extreme points of a signal or data by adding different Gaussian noise with given amplitude into the signal every time, and then performs overall averaging on IMF components obtained by multiple Decomposition to achieve the purpose of offsetting the noise added into the signal, thereby effectively avoiding the appearance of modal aliasing.
The noise amplitude values added to the signal are initialized.
Secondly, decomposing the noise signal added for the ith time:
a. adding a Gaussian signal n with a certain amplitude to the signal x (t)i(t),xi(t)=ni(t) + x (t), where the left side of the equation represents the signal formed after the i-th addition of the noise signal; the first term on the right of the equation represents the gaussian noise signal; x (t) represents the original signal.
b. For xi(t) EMD decomposition to obtain a set of IMF components ci(I ═ 1,2, 3.., I), where c isiRepresenting IMF components obtained by the ith decomposition; i represents the number of decompositions.
c. If I < I, return to step a and let I +1, repeat steps a and b until I.
And calculating the overall average value of the I decomposition times and indicating the overall average value by the symbol y.
And fourthly, saving the total average value of all IMF components as the final IMF component.
(3) Fault diagnosis based on deep learning network
And selecting a Restricted Boltzmann Machine (RBM) as a depth network design model. The most important of these is the regulation of RBM. RBM regulation is regulated by progressive layers, a transmission layer of the whole depth model is formed by using 1 hidden layer and 1 visual output layer, and the weight of the whole network is initialized.
The RBM model is represented as RBM (W, b, c, v)0) Where W is the weight connection matrix between the first hidden layer and the second hidden layer, b is the bias of the hidden layer, c is the bias of the input layer, v0Representing the sample set used in training.
Calculating mapping operation between hidden layers for nodes i of all hidden layersI.e. P (h)0i=1|v0) According to P (h)0i=1|v0) Sampling to obtain h0iSigm () is a mapping function.
Calculating mapping operation between visual layers for nodes j of all visual layersI.e. P (v)1j=1|h0) According to P (v)1j=1|h0) Sampling to obtain v1j。
Calculating the mapping operation between the hidden layer and the hidden layer for the used hidden layer node iI.e. P (h)1i=1|v1) According to P (h)1i=1|v1) And updating the bias parameters of the connection weight. The update formula is as follows:
W=W-ε(h0v0'-Q(h1=1|v1)v1')
b=b-ε(h0-Q(h1=1|v1))
c=c-ε(v0-v1)
(4) detecting faults of turbine blades of a combustion engine: a fault detection system for a turbine blade of a combustion engine for an algorithm experiment is constructed, a series of targeted experiments are carried out by taking the system as a platform, and the system is configured as follows:
hardware: the processor Intel (R) Pentium (R) Dual CPU 1.60 GHz; 2GB of memory; a display card 256M; and a hard disk 80G.
Software: windows XP operating system; matlab2012a development environment.
Simulating 6 types of fault data, 10 groups of samples of each type and 60 groups of data in total, carrying out EEMD decomposition on a simulated turbine blade fault signal, then solving the energy of IMF of the simulated turbine blade fault signal, selecting the first 8 components because fault information is mainly concentrated in the first components, wherein the capacity of residual terms is almost zero, and then solving a feature vector. The data were divided into two parts, 48 groups were randomly taken as training samples, and the remaining 12 groups were taken as test samples. And training the deep learning network by using the training samples. And classifying the samples to be tested by using the trained deep learning network, and outputting a classification result, wherein the accuracy is 91.6%.
Simulating 6 types of fault data, 10 groups of samples of each type and 60 groups of data in total, carrying out EEMD decomposition on a simulated turbine blade fault signal, then solving the energy of IMF of the simulated turbine blade fault signal, selecting the first 8 components because fault information is mainly concentrated in the first components, wherein the capacity of residual terms is almost zero, and then solving a feature vector. The data were divided into two parts, 48 groups were randomly taken as training samples, and the remaining 12 groups were taken as test samples. And training the deep learning network by using the training samples. Classifying the samples to be tested by using the trained deep learning network, outputting the classification result, and performing test comparison with a BP neural network and a Support Vector Machine (SVM), wherein the result is shown in table 1.
TABLE 1 comparison of diagnostic results
Claims (1)
1. A method for detecting faults of turbine blades of a combustion engine based on deep learning is characterized by comprising the following steps:
(1) the temperature data preprocessing of the turbine blade comprises the following steps:
(1.1) periodic Signal extraction
The turbine engine analyzed consisted of 86 blades, the data for radiation thermometry was obtained after a number of cycles of rotation of all the blades,
according to the discriminant formula of the periodic property: f (x) ═ f (x + T)
Wherein T is the period of the function, and whether the temperature data is period data is judged by the above formula; the minimum value and the maximum value are taken as the discrimination standard to carry out the period discrimination,
taking a certain maximum value point as a starting point, counting all the maximum value points in a certain section of interval, calling the points as test points, obtaining two adjacent test points with the same temperature, then respectively sampling backwards and comparing the temperatures of the sampling points, and if the two adjacent test points with the same temperature are the same, determining that the period is one;
(1.2) variable Condition temperature data distribution
The working conditions are divided into the following three working conditions: 0.6 operating mode, 0.8 operating mode and 1.0 operating mode, the turbine blade rotational speed under three kinds of operating modes respectively is: 8635/rpm, 8920/rpm, 9138/rpm;
(1.3) Single blade temperature data distribution
Establishing a segmentation algorithm for single blade temperature data on extreme value searching, counting maximum value points in the data, comparing the temperature data of each extreme value point, selecting a point with the maximum temperature data, namely the maximum value point, and marking as a0, searching backwards for a period, namely 40 sampling points, by taking a0 as a starting point, and recording the maximum value points in the period; judging the maximum value points, when the maximum value points cannot be larger than 10% of the adjacent maximum value points, if the maximum value points are larger than 10% of the adjacent maximum value points, changing the maximum value points into a starting point a0, circulating the process for 86 times, obtaining the period of 86 blades, carrying out sequence numbering on the blades from 01 to 86, and extracting the characteristics of the blades for state classification detection;
(2) extracting turbine blade temperature feature vectors
Extracting the temperature characteristic vector of the blade by adopting a method based on ensemble empirical mode decomposition; changing the number of extreme points and the distribution interval of the extreme points of the signal or data by adding different Gaussian noises with given amplitudes into the signal every time, and then carrying out overall average cancellation on IMF components obtained by multiple decomposition to cancel the noises added into the signal;
(2.1) initializing a noise amplitude value added to the signal;
(2.2) decomposing the ith added noise signal;
(2.2.1) adding a Gaussian signal n with a certain amplitude to the signal x (t)i(t),xi(t)=ni(t) + x (t), where the left side of the equation represents the signal formed after the i-th addition of the noise signal; the first term on the right of the equation represents the gaussian noise signal; x (t) represents the original signal;
(2.2.2) for xi(t) EMD decomposition to obtain a set of IMF components ciWherein c isiRepresenting IMF components obtained by the ith decomposition; i represents the number of decompositions; 1,2,3, ·, I;
(2.2.3) if I < I, returning to step (2.2.1) and so that I ═ I +1, repeating step (2.2.1) and step (2.2.2) until I ═ I;
(2.3) calculating the overall average value by the I decomposition, denoted by the symbol y;
(2.4) saving the total average value of all IMF components as a final IMF component;
(3) fault diagnosis based on deep learning network
Selecting a restricted Boltzmann machine model as a depth network design model, adjusting the restricted Boltzmann machine model through the progressive between layers, forming a transmission layer of the whole depth model by using 1 hidden layer and 1 visual output layer, and initializing the weight of the whole network;
the restricted boltzmann model is represented as RBM (W, b, c, v0), where W is a weight connection matrix between a first hidden layer and a second hidden layer, b is a bias of the hidden layers, c is a bias of the input layers, and v0 represents a sample set used in training;
(3.1) for all nodes i of the hidden layer, calculating mapping operation between the hidden layersI.e. P (h)0i=1|v0) According to P (h)0i=1|v0) Sampling to obtain h0iSigm () is a mapping function;
(3.2) for all nodes j of the visual layers, calculating the mapping operation between the visual layersI.e. P (v)1j=1|h0) According to P (v)1j=1|h0) Sampling to obtain v1j;
(3.3) for the used hidden layer node i, calculating the mapping operation between the hidden layer and the hidden layerI.e. P (h)1i=1|v1) According to P (h)1i=1|v1) Updating the bias parameters of the connection weight; the update formula is as follows:
W=W-ε(h0v′0-P(h1=1|v1)v′1)
b=b-ε(h0-P(h1=1|v1))
c=c-ε(v0-v1);
(4) detecting faults of turbine blades of a combustion engine: constructing a gas turbine blade fault detection system for an algorithm experiment, simulating 6 types of fault data, 10 groups of samples of each type, and 60 groups of data, carrying out EEMD decomposition on a simulated turbine blade fault signal, then solving the energy of IMF of the simulated turbine blade fault signal, selecting the first 8 components, enabling the capacity of a residual item to be zero, and then solving a feature vector; dividing the data into two parts, randomly taking 48 groups as training samples, and taking the other 12 groups as test samples; training the deep learning network by using the training samples; and classifying the samples to be tested by using the trained deep learning network, and outputting a classification result.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610839954.0A CN106248368B (en) | 2016-09-21 | 2016-09-21 | Combustion engine turbine blade fault detection method based on deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610839954.0A CN106248368B (en) | 2016-09-21 | 2016-09-21 | Combustion engine turbine blade fault detection method based on deep learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106248368A CN106248368A (en) | 2016-12-21 |
CN106248368B true CN106248368B (en) | 2019-12-31 |
Family
ID=57600399
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610839954.0A Active CN106248368B (en) | 2016-09-21 | 2016-09-21 | Combustion engine turbine blade fault detection method based on deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106248368B (en) |
Families Citing this family (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107038319B (en) * | 2017-05-16 | 2020-06-16 | 哈尔滨工程大学 | Troubleshooting method for abnormal gear sound of diesel engine generator set |
CN107505931A (en) * | 2017-07-10 | 2017-12-22 | 南京航空航天大学 | A kind of aircraft engine maintenance level decisions method based on depth belief network |
CN107631867B (en) * | 2017-09-07 | 2019-10-15 | 天津工业大学 | A kind of rotating machinery fault intelligent method for classifying based on deep learning |
CN108344564B (en) * | 2017-12-25 | 2019-10-18 | 北京信息科技大学 | A kind of state recognition of main shaft features Testbed and prediction technique based on deep learning |
CN108681747B (en) * | 2018-05-11 | 2021-05-04 | 武汉理工大学 | Rotary machine fault diagnosis and state monitoring system and method based on deep learning |
CN108760813B (en) * | 2018-06-05 | 2020-09-25 | 哈尔滨工程大学 | Gas turbine blade health monitoring system and method based on temperature signals |
CN110608430A (en) * | 2018-06-15 | 2019-12-24 | 广东美的生活电器制造有限公司 | Food processor, steam generator control device, method and readable storage medium |
CN109145134A (en) * | 2018-07-27 | 2019-01-04 | 郑州云海信息技术有限公司 | A kind of method and device of detection device failure |
GB201908496D0 (en) | 2019-06-13 | 2019-07-31 | Rolls Royce Plc | Computer-implemented methods for determining compressor operability |
GB201908494D0 (en) | 2019-06-13 | 2019-07-31 | Rolls Royce Plc | Computer-implemented methods for training a machine learning algorithm |
CN112816200B (en) * | 2021-01-15 | 2022-10-18 | 东方电气集团科学技术研究院有限公司 | Rotary machine axis track identification method based on deep learning |
CN113029533B (en) * | 2021-02-25 | 2022-09-23 | 南京信息工程大学 | Cooling coil fault diagnosis method based on hidden Markov model and discrete P control diagram |
CN113779778B (en) * | 2021-08-24 | 2024-04-05 | 中国舰船研究设计中心 | Ship system task channel test point optimization method |
Family Cites Families (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN100561500C (en) * | 2006-11-16 | 2009-11-18 | 上海交通大学 | Human face posture recognition methods based on limited Boltzmann machine neural network |
CN104636732B (en) * | 2015-02-12 | 2017-11-07 | 合肥工业大学 | A kind of pedestrian recognition method based on the deep belief network of sequence |
CN104915654B (en) * | 2015-06-11 | 2018-06-01 | 浙江工业大学 | A kind of path point data Activity recognition method based on limited Boltzmann machine |
-
2016
- 2016-09-21 CN CN201610839954.0A patent/CN106248368B/en active Active
Also Published As
Publication number | Publication date |
---|---|
CN106248368A (en) | 2016-12-21 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106248368B (en) | Combustion engine turbine blade fault detection method based on deep learning | |
Saufi et al. | Gearbox fault diagnosis using a deep learning model with limited data sample | |
Zhu et al. | Intelligent fault diagnosis of hydraulic piston pump combining improved LeNet-5 and PSO hyperparameter optimization | |
Zhu et al. | Acoustic signal-based fault detection of hydraulic piston pump using a particle swarm optimization enhancement CNN | |
CN111709292A (en) | Compressor vibration fault detection method based on recursive graph and deep convolutional network | |
CN113255432B (en) | Turbine vibration fault diagnosis method based on deep neural network and manifold alignment | |
CN109978039B (en) | Fan blade icing prediction method based on unbalanced data set | |
CN110263834B (en) | Method for detecting abnormal value of new energy power quality | |
Wang et al. | Intelligent fault detection scheme for constant-speed wind turbines based on improved multiscale fuzzy entropy and adaptive chaotic Aquila optimization-based support vector machine | |
CN113469230B (en) | Rotor system deep migration fault diagnosis method, system and medium | |
CN110501683B (en) | Sea-land clutter classification method based on four-dimensional data characteristics | |
CN116754230A (en) | Bearing abnormality detection and fault diagnosis method based on deep convolution generation countermeasure network | |
CN104809478B (en) | A kind of image block method and device towards extensive three-dimensional reconstruction | |
Feng et al. | Fsrf: an improved random forest for classification | |
Xie et al. | Locally generalized preserving projection and flexible grey wolf optimizer-based ELM for fault diagnosis of rolling bearing | |
Wang et al. | Single and simultaneous fault diagnosis of gearbox via wavelet transform and improved deep residual network under imbalanced data | |
Wu et al. | Performance evaluation of adversarial attacks: Discrepancies and solutions | |
Jin et al. | A rotating stall warning method for aero-engine compressor based on DeepESVDD-CNN | |
CN112116101A (en) | Aero-engine fault diagnosis method based on group reduction kernel limit learning machine | |
Hu et al. | Tree species identification method based on improved YOLOv7 | |
CN114444544A (en) | Signal classification and identification method based on convolutional neural network and knowledge migration | |
CN113065133A (en) | Malicious software detection method based on singular spectrum transformation | |
Zhu et al. | Sparse random projection-based hyperdisk classifier for bevel gearbox fault diagnosis | |
Yan et al. | YOLOx model-based object detection for microalgal bioprocess | |
Li et al. | A signal based “W” structural elements for multi-scale mathematical morphology analysis and application to fault diagnosis of rolling bearings of wind turbines |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |